Practices: triggers and modes

A practice acts on the board when something happens to a task (board #2474). What it does is declared, the project has to opt in, every change is attributed to the practice, and an automatic change undoes in one step.

• Trigger: tags on the practice say which events fire it: trigger:created, trigger:enters-inbox, trigger:status:<value>, trigger:tagged:<key>, trigger:closed, trigger:reopened and trigger:commented.
• Procedure: virta procedure #n --when "<filter>" --set kind:bug. The task must match the filter, and it gets the tags.
• Mode: mode:manual never runs itself. mode:suggest proposes the tags on the task, for a person to accept or overrule. mode:automatic sets them itself, recorded so one resolve … undo reverses exactly that.
• Opt-in: virta project set <name> --practices "#12 #40". A practice never touches a project that hasn't opted in, and only acts on what happened after it did.
• Running: virta project practices [name] (and the MCP tool) evaluates the opted-in projects. Every event it writes has an id made from the practice and the event that fired it, so running it twice, or on two machines, writes nothing twice.
• Cascades: a practice's own writes don't fire practices, unless one declares chains-from:#n.

Every outcome (suggested, accepted, overruled, applied, undone) is on the task and counted on the practice, under the measure in force when it happened: the evidence its measure reads (board #2475). The practice's evidence grade moves with that record: decree until it has been applied, observed once it has (or while it has no measure), measured once it has a measure and ten outcomes under it.